SciCombinator

Localization of objects and events in the environment is critical for survival, as many perceptual and motor tasks rely on estimation of spatial location. Therefore, it seems reasonable to assume that spatial localizations should generally be accurate. Curiously, some previous studies have reported biases in visual and auditory localizations, but these studies have used small sample sizes and the results have been mixed. Therefore, it is not clear (1) if the reported biases in localization responses are real (or due to outliers, sampling bias, or other factors), and (2) whether these putative biases reflect a bias in sensory representations of space or a priori expectations (which may be due to the experimental setup, instructions, or distribution of stimuli). Here, to address these questions, a dataset of unprecedented size (obtained from 384 observers) was analyzed to examine presence, direction, and magnitude of sensory biases, and quantitative computational modeling was used to probe the underlying mechanism(s) driving these effects. Data revealed that, on average, observers were biased towards the center when localizing visual stimuli, and biased towards the periphery when localizing auditory stimuli. Moreover, quantitative analysis using a Bayesian Causal Inference framework suggests that while pre-existing spatial biases for central locations exert some influence, biases in the sensory representations of both visual and auditory space are necessary to fully explain the behavioral data. How are these opposing visual and auditory biases reconciled in conditions in which both auditory and visual stimuli are produced by a single event? Potentially, the bias in one modality could dominate, or the biases could interact/cancel out. The data revealed that when integration occurred in these conditions, the visual bias dominated, but the magnitude of this bias was reduced compared to unisensory conditions. Therefore, multisensory integration not only improves the precision of perceptual estimates, but also the accuracy.

Effort and reward jointly shape many human decisions. Errors in predicting the required effort needed for a task can lead to suboptimal behavior. Here, we show that effort estimations can be biased when retrospectively reestimated following receipt of a rewarding outcome. These biases depend on the contingency between reward and task difficulty and are stronger for highly contingent rewards. Strikingly, the observed pattern accords with predictions from Bayesian cue integration, indicating humans deploy an adaptive and rational strategy to deal with inconsistencies between the efforts they expend and the ensuing rewards.

Risk-of-bias assessments are now a standard component of systematic reviews. At present, reviewers need to manually identify relevant parts of research articles for a set of methodological elements that affect the risk of bias, in order to make a risk-of-bias judgement for each of these elements. We investigate the use of text mining methods to automate risk-of-bias assessments in systematic reviews. We aim to identify relevant sentences within the text of included articles, to rank articles by risk of bias and to reduce the number of risk-of-bias assessments that the reviewers need to perform by hand.

Adaptive clinical trials are an innovative trial design aimed at reducing resources, decreasing time to completion and number of patients exposed to inferior interventions, and improving the likelihood of detecting treatment effects. The last decade has seen an increasing use of adaptive designs, particularly in drug development. They frequently differ importantly from conventional clinical trials as they allow modifications to key trial design components during the trial, as data is being collected, using preplanned decision rules. Adaptive designs have increased likelihood of complexity and also potential bias, so it is important to understand the common types of adaptive designs. Many clinicians and investigators may be unfamiliar with the design considerations for adaptive designs. Given their complexities, adaptive trials require an understanding of design features and sources of bias. Herein, we introduce some common adaptive design elements and biases and specifically address response adaptive randomization, sample size reassessment, Bayesian methods for adaptive trials, seamless trials, and adaptive enrichment using real examples.

Single molecule localization microscopy (SMLM) is on its way to become a mainstream imaging technique in the life sciences. However, analysis of SMLM data is biased by user provided subjective parameters required by the analysis software. To remove this human bias we introduce here the Auto-Bayes method that executes the analysis of SMLM data automatically. We demonstrate the success of the method using the photoelectron count of an emitter as selection characteristic. Moreover, the principle can be used for any characteristic that is bimodally distributed with respect to false and true emitters. The method also allows generation of an emitter reliability map for estimating quality of SMLM-based structures. The potential of the Auto-Bayes method is shown by the fact that our first basic implementation was able to outperform all software packages that were compared in the ISBI online challenge in 2015, with respect to molecule detection (Jaccard index).

How we attend to our thoughts affects how we attend to our environment. Holding information in working memory can automatically bias visual attention toward matching information. By observing attentional biases on reaction times to visual search during a memory delay, it is possible to reconstruct the source of that bias using machine learning techniques and thereby behaviorally decode the content of working memory. Can this be done when more than one item is held in working memory? There is some evidence that multiple items can simultaneously bias attention, but the effects have been inconsistent. One explanation may be that items are stored in different states depending on the current task demands. Recent models propose functionally distinct states of representation for items inside versus outside the focus of attention. Here, we use behavioral decoding to evaluate whether multiple memory items-including temporarily irrelevant items outside the focus of attention-exert biases on visual attention. Only the single item in the focus of attention was decodable. The other item showed a brief attentional bias that dissipated until it returned to the focus of attention. These results support the idea of dynamic, flexible states of working memory across time and priority.

Creating large datasets for biomedical relation classification can be prohibitively expensive. While some datasets have been curated to extract protein-protein and drug-drug interactions from text, we are also interested in other interactions including gene-disease and chemical-protein connections. Also, many biomedical researchers have begun to explore ternary relationships. Even when annotated data is available, many datasets used for relation classification are inherently biased. For example, issues such as sample selection bias typically prevent models from generalizing in the wild. To address the problem of cross-corpora generalization, we present a novel adversarial learning algorithm for unsupervised domain adaptation tasks where no labeled data is available in the target domain. Instead, our method takes advantage of unlabeled data to improve biased classifiers through learning domain-invariant features via an adversarial process. Finally, our method is built upon recent advances in neural network methods.

A biomolecular ensemble exhibits different responses to a temperature gradient depending on its diffusion properties. MicroScale Thermophoresis technique exploits this effect and is becoming a popular technique for analyzing interactions of biomolecules in solution. When comparing affinities of related compounds, the reliability of the determined thermodynamic parameters often comes into question. The thermophoresis binding curves can be assessed by Bayesian inference, which provides a probability distribution for the dissociation constant of the interacting partners. By applying Bayesian machine learning principles, binding curves can be autonomously analyzed without manual intervention and without introducing subjective bias by outlier rejection. We demonstrate the Bayesian inference protocol on the known survivin:borealin interaction and on the putative protein-protein interactions between human survivin and two members of the human Shugoshin-like family (hSgol1 and hSgol2). These interactions were identified in a protein microarray binding assay against survivin and confirmed by MicroScale Thermophoresis.

Health economic decision models often involve a wide-ranging and complicated synthesis of evidence from a number of sources, making design and implementation of such models resource-heavy. When new data become available and reassessment of treatment recommendations is warranted, it may be more efficient to perform a Bayesian update of an existing model than to construct a new model. If the existing model depends on many, possibly correlated, covariates, then an update may produce biased estimates of model parameters if some of these covariates are completely absent from the new data. Motivated by the need to update a cost-effectiveness analysis comparing diagnostic strategies for coronary heart disease, this study develops methods to overcome this obstacle by either introducing additional data or using results from previous studies. We outline a framework to handle unobserved covariates, and use our motivating example to illustrate both the flexibility of the proposed methods and some potential difficulties in applying them.

How do people recognize and learn about complex functional structure? Taking inspiration from other areas of cognitive science, we propose that this is achieved by harnessing compositionality: complex structure is decomposed into simpler building blocks. We formalize this idea within the framework of Bayesian regression using a grammar over Gaussian process kernels, and compare this approach with other structure learning approaches. Participants consistently chose compositional (over non-compositional) extrapolations and interpolations of functions. Experiments designed to elicit priors over functional patterns revealed an inductive bias for compositional structure. Compositional functions were perceived as subjectively more predictable than non-compositional functions, and exhibited other signatures of predictability, such as enhanced memorability and reduced numerosity. Taken together, these results support the view that the human intuitive theory of functions is inherently compositional.